ESMO Asia Congress 2022

Abstract 6P

Background

Several studies have indicated that magnetic resonance imaging radiomics can predict survival in patients with breast cancer, but the potential biological underpinning remains indistinct. This study aims to develop an interpretable deep-learning-based network for classifying recurrence risk and revealing the potential biological underpinning.

Methods

In this multicenter study, 1,113 nonmetastatic invasive breast cancer patients were included, and were divided into the training cohort (n=698), the validation cohort (n=171), and the testing cohort (n=244). This study included three phases to develop the Radiomic DeepSurv Net (RDeepNet). In phase 1, the RDeepNet was constructed with deep learning radiomic features for recurrence-free survival (RFS) prediction. RNA-sequencing was performed to explore the mechanisms of radiomics. In phase 2, correlation and variance analyses were conducted to examine changes of radiomics in patients after neoadjuvant chemotherapy. The association and quantitative relation of radiomics and epigenetic molecular characteristics were further analyzed in phase 3.

Results

The RDeepNet was significantly associated with RFS (HR 0.03, 95% CI 0.02–0.06, P < 0.001), and it achieved AUCs of 0.98, 0.94, and 0.92 for 1-, 2-, 3-year RFS. The RDeepNet showed AUCs of 0.91 and 0.94 for 3-year RFS in the validation and testing cohorts, respectively. The RDeepNet was generalized by validation in different molecular subtypes and patients with different therapy regimens (All P < 0.001). The radiomic features were found to vary among patients with different therapeutic responses and after neoadjuvant chemotherapy. Moreover, long non-coding RNAs (lncRNAs) were discovered to be significantly correlated with radiomics and could be quantified by radiomics.

Conclusions

The findings indicate that the RDeepNet can be conveniently used to guide treatment decisions and predict lncRNA expression noninvasively in breast cancer.